Theor Appl Climatol (2014) 115:713–729 DOI 10.1007/s00704-013-0917-x ORIGINAL PAPER Assessing the performance of satellite-based precipitation products and its dependence on topography over Poyang Lake basin Xianghu Li & Qi Zhang & Chong-Yu Xu Received: 3 September 2012 / Accepted: 22 April 2013 / Published online: 26 May 2013 # Springer-Verlag Wien 2013 Abstract Satellite-based precipitation products (SPPs) have greatly improved their applicability and are expected to offer an alternative to ground-based precipitation estimates in the present and the foreseeable future. There is a strong need for a quantitative evaluation of the usefulness and limitations of SPPs in operational meteorology and hydrology. This study compared two widely used high-resolution SPPs, the Tropical Rainfall Measuring Mission (TRMM) and Precipitation Estimation from Remote Sensing Information using Artificial Neural Network (PERSIANN) in Poyang Lake basin which is located in the middle reach of the Yangtze River in China. The bias of rainfall amount and occurrence frequency under different rainfall intensities and the dependence of SPPs performance on elevation and slope were investigated using different statistical indices. The results revealed that (1) TRMM 3B42 usually underestimates the rainy days and overestimates the average rainfall as well as annual rainfall, while the PERSIANN data were markedly lower than rain gauge data; (2) the rainfall contribution rates were underestimated by TRMM 3B42 in the middle rainfall class but overestimated in the heavy rainfall class, X. Li : Q. Zhang (*) State Key Laboratory of Lake Science and Environment, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, 73 East Beijing Road, Nanjing 210008, People’s Republic of China e-mail: qzhang@niglas.ac.cn X. Li e-mail: xhli@niglas.ac.cn C.<Y. Xu Department of Hydrology and Water Resources, Wuhan University, Wuhan, China e-mail: c.y.xu@geo.uio.no C.<Y. Xu Department of Geosciences, University of Oslo, Oslo, Norway while the opposite trend was observed for PERSIANN; (3) although the temporal distribution characteristics of monthly rainfall were correctly described by both SPPs, PERSIANN tended to suffer a systematic underestimation of rainfall in every month; and (4) the performances of both SPPs had clear dependence on elevation and slope, and their relationships can be fitted using quadratic equations. 1 Introduction Precipitation is the key input for hydrological modeling and its temporal and spatial distribution has a significant impact on the land surface hydrological fluxes and states (Gottschalck et al. 2005; Tian et al. 2007; Su et al. 2008). Therefore, accurate measurements of precipitation on fine spatial and temporal scales are very important for simulating land surface hydrologic processes, predicting drought and flood, and monitoring water resources (Sorooshian et al. 2005; Yong et al. 2010). However, in many populated regions of the world and especially in developing countries, groundbased measurement networks (either from rain gauge or weather radar) are either sparse in both time and space or nonexistent (Behrangi et al. 2011), and their limited sampling areas and problems inherent in point measurements represent a substantial difficulty when dealing with effective spatial coverage of rainfall over a large area (Pegram et al. 2004; Schulze 2006; Ghile et al. 2010). Although weather radar has enormous potential to offer rainfall estimates with high spatial resolution and temporal continuity (Sun et al. 2000; He et al. 2011), there is often a large space–time variable bias (Smith et al. 2007; Krajewski and Smith 2002) and its accuracy is highly sensitive to atmospheric conditions, sampling height of the radar beam, beam blocking, 714 variations in the reflectivity–rainfall rate relationships, ground echoes, and distance from the radar (Deyzel et al. 2004; Pegram et al. 2004; Piccolo and Chirico 2005). This situation restricts these regions to manage water resources (Behrangi et al. 2011) and hampers the development and use of flood and drought warning models, extreme weather monitoring, and decisionmaking systems (AghaKouchak et al. 2011). Alternatively, satellite-based precipitation products (SPPs) are widely accepted as promising strategies to address the previously mentioned limitations (Ghile et al. 2010). Such data are especially valuable in developing countries or remote locations, where conventional rain gauge or weather radar data are sparse or of bad quality (Hughes 2006). Furthermore, the near real-time availability of the SPPs makes them suitable for modeling applications where water resources management is crucial and data gathering and quality assurance are cumbersome (Stisen and Sandholt 2010). Recent development in global and regional SPPs has greatly improved their applicability as input to large-scale distributed hydrological models (Stisen and Sandholt 2010; Li et al. 2012, 2013; Samaniego et al. 2012) and are expected to offer an alternative to groundbased rainfall estimates in the present and the foreseeable future (Sawunyama and Hughes 2008). This is mainly due to the increased temporal and spatial resolution of SPPs and also due to improved accuracy resulting from new methods to merge various data sources such as radar, microwave, and thermal infrared (TIR) remote sensing (Gottschalck et al. 2005; Tian and Peters-Lidard 2007; Stisen and Sandholt 2010). With suites of sensors flying on a variety of satellites over the last two decades, many satellite-based precipitation estimation algorithms have been developed (Behrangi et al. 2011) to combine measurements of different spaceborne sensors and gauge data allow the derivation of high-quality precipitation estimates. Since Huffman et al. (1995, 1997) created a scheme to combine satellite data of different sensors (microwave, infrared [IR], and longwave radiation) with gauge data and built the Global Precipitation Climatology Project (GPCP) combined precipitation dataset at a 2.5×2.5° grid and monthly resolution, these algorithms have improved constantly by emerging further multisource products with higher resolutions (Scheel et al. 2011). Currently, several satellite-based gridded precipitation estimates are available for (at least) the lower latitudes and the tropics at high temporal (three hourly or shorter) and reasonably high spatial (0.25×0.25° or finer) resolutions. Examples include the Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) (Huffman et al. 2007), the Climate Prediction Center (CPC) morphing algorithm (CMORPH) (Joyce et al. 2004), the Precipitation Estimation from Remote Sensing Information using Artificial Neural Network (PERSIANN) (Hsu et al. 1997; Sorooshian et al. 2000), the Global Satellite Mapping of X. Li et al. Precipitation (Kubota et al. 2007; Aonashi et al. 2009; Ushio et al. 2009), the Naval Research Laboratory Global BlendedStatistical Precipitation Analysis (Turk et al. 2000), and so on. Although different in the precipitation estimation procedure, in all of the listed products, combined information from passive microwave (PMW) sensors in low earth orbiting satellites and IR radiometers in geostationary earth orbiting (GEO) satellites is used to improve the consistency, accuracy, coverage, and timeliness of high-resolution precipitation data (Kubota et al. 2009; Behrangi et al. 2011). However, satellite data also suffer from some inherent shortcomings and have biases and random errors that are caused by various factors like sampling frequency, nonuniform field of view of the sensors, and uncertainties in the rainfall retrieval algorithms (Nair et al. 2009). It is, therefore, essential to validate the satellite-derived products with conventional rain estimates to quantify the direct usability of these products (Nair et al. 2009; Li et al. 2012, 2013). Numerous researchers have examined the quality of satellite-derived precipitation datasets in various regions of the world. Table 1 summarizes recent studies on evaluations of three widely used high-resolution SPPs (TMPA, CMORPH, and PERSIANN) based on the work of Romilly and Gebremichael (2011). For instance, Dinku et al. (2010) evaluated two satellite rainfall estimation algorithms, TMPA and CMORPH, over two locations (highlands of Ethiopia and Columbia) and found that total rainfall amount was overestimated by TMPA 3B42RT (13 %) and CMORPH (11 %) in Ethiopia, while it was underestimated by TMPA 3B42RT (17 %), TMPA 3B42 (16 %), and CMORPH (9 %) in Columbia. Stisen and Sandholt (2010) evaluated five satellite products, including TRMM 3B42, CMORPH, and PERSIANN, in the Senegal River Basin using the MIKE SHE hydrological model and found that TRMM 3B42 performs better than the other satellite products. Yamamoto et al. (2011) compared several SPPs with rainfall data from the automated weather station in the Nepal Himalayas and found that PERSIANN showed large differences with the observed values in winter and CMORPH had a tendency to overestimate precipitation in the pre-monsoon and postmonsoon seasons. Ward et al. (2011) believed that TRMM 3B42 and PERSIANN are unable to detect light rainfall amounts and underestimate rainfall in the dry season. Behrangi et al. (2011) found that TMPA 3B42RT, CMORPH, and PERSIANN tend to overestimate intense precipitation during warm months. In addition, several studies have also validated the effects of topography on SPPs performance. For example, Bitew and Gebremichael (2010) found that the CMORPH and PERSIANN-Cloud Classification System (CCS) underestimated 32 and 49 % of total rainfall, respectively, in a high-elevation region. Hong et al. (2007) evaluated the impact of topography on the performance of PERSIANN-CCS in western Mexico and found Assessing the performance of satellite-based precipitation products 715 Table 1 Evaluations on high-resolution SPPs (supplement based on Romilly and Gebremichael 2011) Precipitation Regions products Main results References TMPA 3B42 TMPA 3B42RT CMORPH PERSIANN TMPA 3B42 CMORPH PERSIANN TRMM 3B42 CMORPH PERSIANN TRMM 3B42 PERSIANN TMPA 3B42RT CMORPH PERSIANN TRMM 3B42 CMORPH PERSIANN TMPA 3B42 TMPA 3B42RT CMORPH TMPA 3B42 TMPA 3B42RT CMORPH TMPA 3B42RT CMORPH PERSIANN TMPA 3B42 CMORPH PERSIANN CMORPH PERSIANNCCS Illinois River basin, USA TMPA 3B42RT, CMORPH, and PERSIANN tend to overestimate intense precipitation during warm months; CMORPH demonstrates higher skill to delineate precipitation area Behrangi et al. (2011) Nepal Himalayas PERSIANN showed large differences in winter; CMORPH overestimates rainfall Yamamoto et al. in the pre-monsoon and post-monsoon seasons; TMPA 3B42 (2011) and CMORPH increase rainfall during the morning Congo River basin TRMM 3B42 provides the best spatial and temporal distributions and magnitudes; Beighley et al. CMORPH and PERSIANN tend to overestimate magnitudes (2011) Paute River, Ecuador and Baker basin, Patagonia Ward et al. (2011) Awash River basin, Ethiopia Both are unable to detect light rainfall amounts and underestimated in the dry season; there is a systematic underestimation of rainfall occurrence by TRMM 3B42 3B42RT and CMORPH show an increasing trend with elevation; PERSIANN considerably underestimates rainfall in high-elevation areas Senegal River Basin, West African CMORPH and PERSIANN have much larger biases than TRMM based on MIKE SHE hydrological model Stisen and Sandholt (2010) Western Highlands, Ethiopia Occurrence of rain underestimated by all products; total amount underestimated by Dinku et al. TMPA 3B42 (14 %) and overestimated by TMPA 3B42RT (13 %) and (2010) CMORPH (11 %) Highlands, Columbia Occurrence of rain underestimated by all products; Total amount underestimated Dinku et al. by TMPA 3B42RT (17 %), TMPA 3B42 (16 %), and CMORPH (9 %) (2010) Great Rift Valley, Ethiopia TMPA 3B42RT and CMORPH show elevation-dependent trends, with underestimation at higher elevations; PERSIANN underestimates at higher elevations and not exhibit elevation-dependent trends Hirpa et al. (2010) Hirpa et al. (2010) USA and the Pacific Ocean CMORPH and PERSIANN overestimate rainfall as much as 125 % in warm season over the USA; CMORPH and TMPA 3B42 underestimate rainfall over the Pacific Ocean Sapiano and Arkin (2009) Berressa basin, Ethiopia Both underestimate heavy rainfall by 50 %; total amount underestimated by CMORPH (32 %) and PERSIANN-CCS (49 %) Bitew and Gebremichael (2010) Overestimate rainfall <1 mm/day; underestimate rainfall >1 mm/day Yu et al. (2009) TMPA 3B42 Mainland China CMORPH TRMM Sierra Madre Occidental, 3B42 Mexico CMORPH PERSIANN TRMM Ethiopia and Zimbabwe 3B42 TRMM 3B42RT CMORPH PERSIANN CMORPH and PERSIANN overestimate the rainfall rate and frequency; TRMM Nesbitt et al. 3B42 closely agrees with the rain gauge network (2008) All data detected the occurrence of rainfall well, the amount of rainfall was poorly estimated; the performance was better over Zimbabwe (relatively flat area) as compared with Ethiopia (complex terrain) Dinku et al. (2008) 716 X. Li et al. Table 1 (continued) Precipitation Regions products Main results TMPA 3B42 Continental USA CMORPH TMPA 3B42 has near zero biases for both summer and winter months; CMORPH Tian et al. (2007) overestimates rainfall over central; underestimates over the northeast during the summer PERSIANN overestimate total rainfall over central and western during Gottschalck et al. spring and summer, underestimate during fall and winter; TMPA 3B42RT (2005) overestimate during spring and summer, overestimate during fall and winter TMPA 3B42RT PERSIANN Continental USA that light precipitation events were underestimated in the high-elevation regions and precipitation events in the lowerelevation regions were overestimated. Hirpa et al. (2010) also found elevation-dependent trends of performance in the TMPA 3B42RT and CMORPH products. Many researchers have testified that the accuracy of SPPs is influenced by location, season, rain type (i.e., convective, stratiform), topography, climatological factors, and so on (Artan et al. 2007; Dinku et al. 2008; Jiang et al. 2008; Han et al. 2011). However, very few of previous studies have so far fully and comprehensively analyzed these aspects. Their performances under different rainfall intensities and in different seasons are still unclear. Moreover, although several recent studies (i.e., Dinku et al. 2008; Ward et al. 2011; Romilly and Gebremichael 2011) have taken into account the effects of elevation, they have only compared the performances of SPPs in high-elevation and lowelevation regions and the quantitative relationships between accuracy and elevation are not mentioned. On the other hand, slope as another important topographic factor is not considered yet in these corresponding studies; the relationship between SPPs performance and slope is still unclear. Poyang Lake, located in the middle reach of the Yangtze River (Fig. 1), is the largest freshwater lake in China and plays a crucial role in flood protection for the lower reaches of the Yangtze River. It has recently been shown that the frequency and severity of the floods have increased since 1990 (Guo et al. 2008) and the surface runoffs from the five subbasins have been the primary source of the major floods in the Poyang Lake basin (Hu et al. 2007). To implement flood protection and regulation and ensure water safety in areas around the lake, it is necessary to understand the flood development and the rainfall–runoff processes in the catchment. However, the applications of satellite-based precipitation, as complementary rainfall data, are seldom in Poyang Lake basin. The scarcity on the accuracy evaluation of SPPs in this basin has hampered their extensive application and development of flood warning models to a certain extent. Therefore, the objectives of the study are designed to evaluate and compare two high-resolution SPPs (TRMM 3B42 and PERSIANN) with rain gauge data and investigate their spatial and temporal characteristics in the Poyang Lake References basin. Also, the bias of rainfall amount and occurrence frequency under different rainfall intensities in each month and the quantitative relationships of accuracy with elevation and slope are investigated. By doing so, different statistical measures and methods are calculated and used in the study. The study is expected to serve as useful reference and valuable information for future study and application of satellite rainfall data in the Poyang Lake basin as well as in other regions. The rest of this paper is organized as follows. In the next section, details of the study area and climate, along with a brief discussion on the rain gauge and SPPs, are presented. In Section 3, the indexes and methods used in the study are briefly described with the help of cited references. Major results of this study are presented in Section 4. Section 5 mainly discusses the possible sources of errors of SPPs from various aspects and the further challenges that we face in using SPPs for hydrological studies, and Section 6 summarizes the conclusions. 2 Study area and data 2.1 Study area Poyang Lake basin is located in the middle and lower reaches of the Yangtze River, China and the lake receives water flows mainly from the five rivers: Xiushui River, Ganjiang River, Fuhe River, Xinjiang River, and Raohe River and discharges into the Yangtze River through a channel in its northern part (Fig. 1). The total drainage area of the water systems is 16.22×104 km2, accounting for 9 % of the drainage area of the Yangtze River basin. The topography in the basin varies from highly mountainous and hilly areas (with the maximum elevation of 2,200 m above mean sea level) to alluvial plains in the lower reaches of the primary watercourses. Poyang Lake basin has a subtropical wet climate characterized with a mean annual precipitation of 1,680 mm for the period of 1960–2007 and annual mean temperature of 17.5 °C. Annual precipitation shows a wet season and a dry season and a short transition period in between. Precipitation increases quickly from January to Assessing the performance of satellite-based precipitation products June and decreases sharply in July, and after September, the dry season sets in and lasts through December. In response to the annual cycle of precipitation, the Poyang Lake can expand to a large water surface of 3,800 km2 and volume of 320×108 m3 in the wet season, but shrinks to little more than a river during the dry season (Xu et al. 2001) and exposes extensive floodplains and wetland areas. 2.2 Data 717 averaged to obtain the areal daily precipitation for the Poyang Lake basin, and the spatial distribution of annual rainfall is interpolated by the inverse distance weighted (IDW) technique with a power of 2. In addition, the digital elevation model data are derived from the National Aeronautics and Space Administration (NASA) Shuttle Radar Topographic Mission at a spatial resolution of 90 m (http://srtm.csi.cgiar.org), which are used to obtain the altitude of each rain gauge and the average slope in pixel size of 0.25×0.25°. 2.2.1 Ground data 2.2.2 Satellite data Daily precipitation data, during the period 2000–2007 for 34 stations in the Poyang Lake basin, are obtained from National Meteorological Information Center of China, which are used to compare and evaluate the accuracy of satellite-based rainfall data in the study. The distribution of rain gauges is shown in Fig. 1. These data have been widely used for different studies previously and the qualities have been approved to be reliable (Hu et al. 2007; Guo et al. 2008; Li et al. 2012). Daily precipitation data from all the stations are Fig. 1 Location of Poyang Lake basin and the distribution of rain gauges (black squares represent the six selected 0.25× 0.25° grids for statistical comparison) The high-resolution SPPs investigated in this study are TRMM 3B42 and PERSIANN. TRMM was launched in November 1997 as a joint effort by NASA and the Japan Aerospace Exploratory Agency with the specific objectives of studying and monitoring the tropical rainfall (Kummerow et al. 1998). The TRMM includes a number of precipitation-related instruments, such as a precipitation radar, a visible and IR sensor, and a special sensor microwave imager (SSM/I) like the TRMM 718 X. Li et al. microwave imager (TMI) (Kummerow et al. 2001), and detailed information is shown in Table 2. Several algorithms have been developed to make use of data from the TRMM mission, and the purpose of the 3B42 class of algorithm is to produce TRMM-adjusted merged IR precipitation and root mean square precipitation error estimates. The TRMM 3B42 precipitation product was produced using the following four steps (Vila et al. 2009). The first stage of the algorithm consists of the calibration and combination of microwave precipitation estimates. Passive microwave observations from the TMI, Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E), and SSM/I are converted to precipitation estimates at the TRMM Science Data and Information System with sensor-specific versions of the Goddard profiling algorithm (Kummerow et al. 2001). In the second step, the IR precipitation estimates are created using the calibrated microwave precipitation. Histograms of time–space matched combined microwave (high-quality precipitation rates) and IR brightness temperatures (TBs), each represented on the same three hourly 0.25×0.25° grid, are accumulated for 1 month into histograms on a 1×1° grid and aggregated to overlapping 3×3° windows, which are then used to create spatially varying calibration coefficients that convert IR TBs to precipitation rates (Huffman et al. 2007; Vila et al. 2009). In the third stage, the microwave and IR estimates are combined. The physically based combined microwave estimates are taken “as is” where available, and the remaining grid boxes are filled with microwave-calibrated IR estimates. And the final step is the indirect use of rain gauge data. The GPCP monthly rain gauge analysis data developed by the Global Precipitation Climatology Center and the Climate Assessment and Monitoring System monthly rain gauge analysis data developed by the CPC are integrated using a histogram-matching technique (Huffman et al. 2007). A detailed description of this algorithm can be found in Huffman et al. (2007) and Dinku and Anagnostou (2006). The PERSIANN dataset, from University of California, Irvine, uses an adaptive neural network function classification/approximation procedure to estimate rainfall rates at each 0.25×0.25° pixel of the IR TB image provided by high-frequency (48 readings a day) geostationary satellites (Geostationary Operational Environmental Satellites (GOES)-8, GOES-9 and GOES-10; Geostationary Meteorological Satellite-5, Meteorological Satellite (MetSat)-6 and MetSat-7) (Hsu et al. 1997; Sorooshian et al. 2000; Asadullah et al. 2008). Model parameters are regularly updated using rainfall estimates from low-orbit satellites, including the TRMM, the National Oceanic and Atmospheric Administration (NOAA)-15, NOAA-16, and NOAA-17 satellites, and the Defense Meteorological Satellite Program (DMSP) F-13, DMSP F-14, and DMSP F-15 satellites (Ferraro and Marks 1995; Kummerow et al. 1998; Hsu and Sorooshian 2008). An adaptive training feature facilitates updating of the network parameters whenever independent estimates of rainfall are available. Initially, the neural network was trained using radar data and the input was limited to TIR data and later extended to include the use of both daytime visible imagery (Hsu et al. 1999) and the TMI rainfall estimates (Sorooshian et al. 2000). In the operation of PERSIANN, two PERSIANN algorithms are running in parallel (Hsu and Sorooshian 2008): one is run in the simulation mode and the other in the update mode. The simulation mode generates the surface rain rate at the 0.25×0.25° resolution at every 30 min from the GEO satellites IR images, while the update mode continuously adjusts the mapping function parameters of PERSIANN based on the fitting error of any pixel for which a PMW instantaneous rainfall estimate is available. The simulation mode generates the regular rainfall rate output, and the update mode improves the quality of the product. A full description of the algorithm was given by Sorooshian et al. (2000) and Hsu et al. (1997). TRMM 3B42 and PERSIANN precipitation estimates are available at the 0.25×0.25° grid, three hourly and six hourly resolution, respectively, with global coverage between 50° N and 50° S, and the data used in the study cover the period from January and March 2000, respectively, to December 2007. Table 2 Summary of the high-resolution SPPs Datasets Spatial coverage Temporal coverage Spatial resolution Temporal Main product data sources resolution TRMM 3B42 V6 50° S–50° N January 1998– 0.25×0.25° 3 hourly globally present PERSIANN 50° S–50° N March 2000– 0.25×0.25° 6 hourly globally present References Geostationary IR, TRMM TMI, SSM/I, Huffman et al. (2007) AMSU, AMSR-E, and rain gauge data Neural network using geostationary IR, Hsu et al. (1997, 1999); TRMM TMI, SSM/I, and AMSU Sorooshian et al. (2000) IR infrared, SSM/I special sensor microwave/imager, AMSU advanced microwave sounding unit, AMSR-E Advanced Microwave Sounding Radiometer for the Earth Observing System Assessing the performance of satellite-based precipitation products Table 3 Contingency table for comparing SPPs with rain gauge data 3 Methods To quantitatively compare SPPs with rain gauge observations, several widely used validation statistical indices are selected in the study. The correlation coefficient (R) is used to reflect the degree of linear correlation between satellite-based precipitation and gauge observations, the mean error (ME) simply scales the average difference between the satellite-based estimates and observations, the root mean square error (RMSE) measures the average error magnitude but gives greater weight to the larger errors, and the relative bias (BIAS) is used to assess the systematic bias of satellite precipitation. The values of R, ME, RMSE, and BIAS are calculated, respectively, as Eqs. 1, 2, 3 and 4: n P Gi G Si S i¼1 R ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n n 2 2 P P Gi G Si S i¼1 i¼1 ð1Þ ME ¼ 719 n 1X ð Si G i Þ n i¼1 ð2Þ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n 1X ð Si G i Þ 2 RMSE ¼ n i¼1 ð3Þ Satellite-based rainfall data Greater than or equal to the threshold Less than the threshold BIAS ¼ i¼1 ð Si G i Þ n P 100 % ð4Þ Less than the threshold a b c d The FAR measures the fraction of rain detections that was actually false alarms. It ranges from 0 to 1, with a perfect score of 0. The ETS provides the fraction of rain events (observed and/or detected) which was correctly detected, and the perfect score is 1 (Su et al. 2008; Koo et al. 2009). The ETS is commonly used as an overall skill measure by the numerical weather prediction community, whereas the FBI, POD, and FAR provide complementary information about bias, misses, and false alarms (Koo et al. 2009). Those indices have been successfully applied in many studies (Layberry et al. 2006; Ebert et al. 2007; Su et al. 2008; Kubota et al. 2009; Yong et al. 2010; Shrestha et al. 2011) and are believed to be robust and provide a sound basis for the assessment of the rainfall detection capabilities of the satellite products. For a more detailed explanation of FBI, POD, FAR, and ETS, please refer to Wilks (2006) and Ebert et al. (2007) and their values are calculated, respectively, using Eqs. 5, 6, 7, 8, and 9: aþb aþc ð5Þ POD ¼ a aþc ð6Þ FAR ¼ b aþb ð7Þ ETS ¼ a He a þ b þ c He ð8Þ Gi i¼1 where n is number of samples, Gi are gauge observations, Si are satellite-based precipitation, and G and S are mean gauge and satellite-based precipitation, respectively. In addition, evaluation and comparison are carried out by detecting rain events at different precipitation thresholds over the Poyang Lake basin at a daily time step. It is performed by computing the frequency bias index (FBI), probability of detection (POD), false alarm ratio (FAR), and equitable threat score (ETS) (Wilks 1995, 2006) based on a 2×2 contingency table, as shown in Table 3. The FBI indicates whether the dataset tends to underestimate (FBI<1) or overestimate (FBI>1) rain events, and it ranges from 0 to infinity, with a perfect score of 1. The POD gives the fraction of rain occurrences that was correctly detected. It ranges from 0 to 1, with a perfect score of 1. Greater than or equal to the threshold a the number of observed rain events correctly detected, b the number of false alarms (rainfall events detected but not observed), c the number of observed rain events not detected, d the sum of cases when neither observed nor detected rain events occurred FBI ¼ n P Rain gauges data He ¼ ð a þ bÞ ð a þ c Þ N ð9Þ where N is the total number of estimates, a is the number of observed rain events correctly detected, b is the number of false alarms (rainfall events detected but not observed), c stands for the number of observed rain events not detected, and d is the sum of cases when neither observed nor detected rain events occurred. 1,527 1,682 1,573 1,796 1,771 1,594 1,657 1,585 1,990 1,430 1,640 1,718 1,349 1,619 152.37 117.8 149.2 152.4 163.3 189.2 154.0 265.5 187.8 222.8 333.5 286.5 511.2 301.2 361.1 258.2 233.5 290.1 280.4 274.2 282.9 62.3 46.6 54.1 72.7 70.1 80.5 64.4 149.9 120.5 106.0 147.4 134.6 304.1 160.4 222.0 152.4 149.7 136.7 156.4 129.5 157.8 8.8 8.2 8.8 9.4 9.2 8.2 8.8 18.3 17.2 15.8 17.2 18.7 17.1 17.4 14.5 13.9 13.4 14.7 14.5 13.8 14.1 Average rainfall in rainy day b a Days with rainfall ≥1 mm 95 89 88 96 97 99 94 83 97 99 104 94 93 95 TRMM Gauge PERSIANN TRMM Gauge PERSIANN TRMM Gauge PERSIANN TRMM Gauge Gauge Max. daily rainfall (mm/day) Average rainfallb (mm/day) Rainy daya (day/year) 108 141 105 110 117 97 113 Gaoan Jinggang Ganzhou Nancheng Yiyang Duchang Average Figure 2 shows the intensity distributions of daily rainfall in different classes and their contributions to the total rainfall Grids 4.2 Evaluation of the rainfall data under different rainfall intensity Table 4 Comparison of statistical indexes between satellite-based and rain gauge rainfall For the comparison between satellite-based rainfall and rain gauge data at grid scale, several statistical indices such as average rainy day in a year, average rainfall in rainy days, maximal daily rainfall, maximal 5-day rainfall, and average annual rainfall were firstly analyzed. Considering the relative location of rain gauges in the grid (central is best), representative of Poyang Lake basin’s topography and the spatial distribution in the catchment, the six grids were selected for the comparison between satellite pixel (0.25×0.25° grid) and the gauging stations inside the grids (namely, Gaoan, Jinggang, Ganzhou, Nancheng, Yiyang, and Duchang) (see Fig. 1) and the results were shown in Table 4. It is seen that the average rainy days (rainfall ≥ 1 mm/day) were 97– 141 days/year for different rain gauges, but 83–104 and 88– 99 days/year for TRMM 3B42 and PERSIANN, respectively. This indicated that average rainy days were underestimated by both SPPs. The average rainfall (in rainy days) is another important and useful index to reflect the precision of rainfall amount. The average rainfalls estimated from rain gauge data ranged between 13.4 and 14.7 mm/day, with an average of 14.1 mm/day. However, the average rainfalls from TRMM 3B42 data were larger than those from rain gauge data in every grid, and the opposite was true for PERSIANN. As for the maximal daily rainfall, the TRMM 3B42 data were smaller than rain gauge data, except in Nancheng and Duchang grid, while PERSIANN data were acutely smaller than rain gauge data in every grid. The similar situations were observed further in the comparison of maximal 5-day rainfall. It is shown that the maximal 5-day rainfall from PERSIANN data were lower markedly than that from rain gauge data, but the performance of TRMM 3B42 was acceptable. Lastly, the average annual total rainfall estimated from rain gauge, TRMM 3B42, and PERSIANN data were 1,619, 1,657, and 839 mm, respectively. TRMM 3B42 overestimated the annual rainfall slightly, but PERSIANN underestimated it greatly. Max. 5-day rainfall (mm/5day) 4.1 Grid-based statistical comparison TRMM 4 Results PERSIANN Annual rainfall (mm/year) All of these indicators are calculated based on the domainaveraged precipitation amount over Poyang Lake basin. Moreover, to quantify the ability of each dataset in predicting light and heavy rainfall events, the FBI, POD, FAR, and ETS are calculated for precipitation thresholds of 1, 2, 5, 10, 25, and 50 mm/day, respectively. 853 744 787 913 908 827 839 X. Li et al. PERSIANN 720 Assessing the performance of satellite-based precipitation products in different grids. It is seen that nonrainy and puny rain (<1 mm/day) had the largest occurrence, occurring about 70 % of the total days, in all datasets. Difference between satellite-based data and rain gauge data was also quite small, except in Jinggang grid that was located in a mountainous area which may bring uncertainties to the observation. The occurrence of the small rainfall class ranges (1 mm/day< rainfall ≤3 mm/day) from TRMM 3B42 was lower than those from rain gauge data, but it was larger for PERSIANN. It can also be seen that, although the occurrences of the first two classes, i.e., nonrainy/puny rain and small rain classes, accounted for as high as 70–80 % of the total days, their contributions to the total rainfall amount were very small. The occurrence of the middle rainfall class 721 ranges (3 mm/day < rainfall ≤ 25 mm/day) estimated by TRMM 3B42 and PERSIANN were generally equivalent (accounting for about 17 % on average) to that of rain gauge data, but with different contribution rates to the total rainfall. For the TRMM 3B42, rainfall contribution rates were mildly lower than that of rain gauge rainfall in classes of 3–10 and 10–25 mm/day. But for the PERSIANN, the contribution rates were overestimated in both classes. It is important to note that the high rainfall ranges (>25 mm/day) play a significant role in contributing to the total rain amount. This kind of information is essential because thunder showers cause the geographical slides and flash floods and hence threaten the economy and human life. Although the two high rainfall classes (25–50 and >50 mm/day) Fig. 2 Distribution of daily rainfall in different rainfall classes and their relative contributions to the total rainfall in different grids 722 occurred only about 5 % of the total days together, their contributions to the total rainfall were as high as 30 and 22 %, respectively, for rain gauge data. TRMM 3B42 performed perfectly for both occurrence and contribution rates in the rainfall class of 25–50 mm/day and the statistics matched well with their counterparts in every grid. However, in the rainstorm class (>50 mm/day), rainfall contribution rates were larger than that of observation data. PERSIANN obviously underestimated both occurrence and contribution rates for the high rainfall ranges, especially for the class of >50 mm/day. So, the SPPs had difficulties in accurately estimating the rainstorm in Poyang Lake basin, TRMM 3B42 inclined to overestimate the occurrence and contribution rates, whereas PERSIANN usually underestimated them. In order to further elucidate the differences between the two datasets, a rain event detection analysis over the Poyang Lake basin had also been performed. Figure 3 shows the verification results of FBI, POD, FAR, and ETS scores for domainaveraged daily precipitation at different rainfall thresholds. It is found that the TRMM 3B42 tended to overestimate the frequency of intense rain events slightly, but there was a systematic underestimation of precipitation occurrence by PERSIANN, as shown in Fig. 3a. The FBI values of the latter decreased from 0.7 to 0.1 as the precipitation threshold increases, indicating that the PERSIANN products were less skillful to correctly capture the magnitude of intense rain events. Figure 3b shows that the POD of both SPPs had a consistent trend and decreased as the precipitation threshold increases, and the POD values of TRMM 3B42 tended to be higher than those of the PERSIANN products. The TRMM 3B42 produced a fine result, with POD values being larger than 0.70 for thresholds 1–5 mm/day, while these values decreased rapidly (POD < 0.5) for thresholds of 25 and Fig. 3 Precipitation detection of daily average TRMM 3B42 and PERSIANN data versus rain gauges at different rainfall thresholds (a FBI, b POD, c FAR, d ETS) X. Li et al. 50 mm/day. Oppositely, the FAR results (Fig. 3c) show an increasing trend as the precipitation threshold increases. TRMM 3B42 and PERSIANN had an equivalent performance with the small FAR scores up to the threshold of 10 mm/day, but the FAR of the latter increased rapidly (>0.7) for precipitation threshold >25 mm/day. The ability to detect rain events was also evaluated in terms of the ETS. Both SPPs showed increasing ETS scores for the precipitation thresholds up to 2 mm/day; then, the ETS scores started dropping for the higher thresholds (Fig. 3d). On the other hand, the ETS scores of TRMM 3B42 were higher than PERSIANN in all precipitation thresholds. Figure 3d, together with Fig. 3b, c, demonstrated that SPPs (TRMM 3B42 and PERSIANN) can identify the small rain events but failed to capture the intense rain events, especially for PERSIANN precipitation products. 4.3 Temporal characteristics of SPPs performance The distribution of monthly rainfall for various precipitation estimates was summarized using box plot for the mean, upper and lower quartiles, and max and min of rainfall as shown in Fig. 4. From the rain gauge data, we can see that the rainfall of Poyang Lake basin increased very fast from January and reached its peak from April to June, then the rainfall decreased sharply from July to September and the dry season set in and lasted through December. Both TRMM 3B42 and PERSIANN data described this distribution characteristic correctly, but the latter tended to suffer from a systematic underestimation of monthly rainfall, regardless of maximum, minimum, or mean. In Fig. 5 several statistical indices, such as RMSE, ME, and BIAS, of monthly averaged daily rainfall between satellitebased and rain gauge data are shown. It is noticeable from Assessing the performance of satellite-based precipitation products 723 Fig. 4 Box plot of monthly rainfall for different precipitation products Fig. 5a that RMSEs ranged from 4 to 10 mm and showed similar temporal pattern for both SPPs. Greater RMSEs were mainly observed in the wet season (April to June), but the lower values were mainly observed in the dry season (October to December). However, the MEs of two SPPs presented the opposite structure (as Fig. 5b). TRMM 3B42 showed positive errors in general and ME values were <1 mm, while the negative errors were mainly found in PERSIANN with large ME values even more than −3 mm. Like ME, BIAS of both SPPs also showed a similar temporal pattern. PERSIANN had a better performance in the summer season with the smaller BIAS (almost 0 in July), but it performed much worse in the winter season with the BIAS value as much as −80 % (Fig. 5b). 4.4 Spatial characteristics of SPPs performance A spatial performance analysis was adopted to examine and compare the spatial variability of SPPs. Figure 6a–c shows the spatial distribution of averaged annual rainfall for the 2000–2007 period derived from rain gauges, TRMM 3B42, and PERSIANN, respectively. The spatial distribution of rain gauge data was directly interpolated by the IDW technique with a power of 2. The rainfall characteristics varied strongly in different areas (Fig. 6a). The largest annual rainfall occurred in the eastern part of Poyang Lake basin (with annual rainfall as high as 2,000 mm), while the lowest one was observed in the southern and northern parts (about 1,400 mm). The median rainfall (about 1,600–1,800 mm) was observed in other areas, i.e., the central parts of the basin. Overall, good agreement existed between the SPPs and rain gauge estimates in terms of relative values within the basin. Both TRMM 3B42 and PERSIANN showed higher Fig. 5 Annual distribution of a RMSE and b ME and BIAS between satellite-based and rain gauge data rainfall rates in the eastern side of the basin than at the central and western sides, although the high rainfall rates covered a larger area than that of the rain gauge estimates (Fig. 6b, c). However, absolute values varied considerably from one dataset to another. For instance, the northeast–northwest rainfall gradient observed in rain gauge estimates, as shown in Fig. 6a, was not reproduced in satellite-based estimates, and local median rainfall in the central parts of the basin was weakened in TRMM 3B42 and PERSIANN. Visual inspections of the results also revealed that the lowest rainfall amount (about 1,050 mm for TRMM 3B42 and 770 mm for PERSIANN) and their distribution regions differ from that of rain gauge estimates, especially for PERSIANN. These biases maybe caused by two aspects: on one hand, there was a weakness for both SPPs to accurately reflect the spatial distribution of precipitation in some regions; on the other hand, a great deal of uncertainties was also exhibited in the spatial distribution of rain gauge data due to the sparsity and bad quality of rain gauges in mountain area as well as the weakness of the interpolation technique. Subsequently, the spatial distribution of R, ME, RMSE, and BIAS between satellite-based and rain gauge data was also examined and compared, which was calculated for every rain gauge and their nearest satellite pixel (0.25× 0.25° grid) during the period of 2000–2007 at the daily scale. The results are shown in Fig. 7. The TRMM 3B42 correlated best with the rain gauge observations, with most R values >0.45 (even >0.55). While for PERSIANN, R values mainly ranged between 0.35 and 0.45. The ME values varied considerably in the two SPPs. TRMM 3B42 showed smaller MEs in the central parts of the basin, with the ME falling into the −0.5 to 0.5 mm class; only several 724 X. Li et al. Fig. 6 Spatial distribution of average annual rainfall for the 2000–2007 period derived from a rain gauges, b TRMM 3B42, and c PERSIANN pixels in the peripheral area produced small positive errors (0.5–1.5 mm). However, the PERSIANN products presented large negative errors in all calculated pixels. Evaluated from the index of the RMSE, PERSIANN performed better than TRMM 3B42. The RMSE values of the former (about 11 mm) were lower than the latter (about 12–14 mm). As for BIAS, its spatial distribution was similar to ME, TRMM 3B42 performed better with smaller relative bias than PERSIANN in general. 4.5 Dependence of SPPs performance on elevation and slope This study also investigated the dependence of SPPs performance on elevation and slope. In order to do so, statistical Fig. 7 Spatial distribution of R, ME, RMSE, and BIAS between satellite-based and rain gauge data Assessing the performance of satellite-based precipitation products 725 Fig. 8 Scatter plots of the correlation coefficient and RMSE versus elevation over the Poyang Lake basin indices such as the correlation coefficient and RMSE between satellite-based and rain gauge data were used, and for elevation, a logarithm transformation (ln(Elevation), simply denoted by ln(E)) was made to obtain a better fit at the lowest values. Figure 8 shows the scatter plots of the correlation coefficient and RMSE versus ln(E) over the Poyang Lake basin. It is seen that the correlation coefficient as well as RMSE has a clear dependence on elevation in both SPPs, and their relationship can be fitted using quadratic equations. The correlation coefficient, either from TRMM 3B42 or PERSIANN, reached maximum at approximately ln(E)= 4.5, and then started dropping for higher ln(E). The RSME values for TRMM 3B42 decreased with increasing ln(E) at lower elevation (ln(E)<4.5) and increased after that. The RMSE from PERSIANN had a feeble dependence on ln(E) at lower elevation and increased clearly at higher elevation. The scatter plots of the correlation coefficient and RMSE versus slope are shown in Fig. 9. It is found that the correlation coefficient and RMSE varied with the slope in both SPPs, and their relationships can also be fitted using quadratic equations. Similar with Fig. 8, the correlation coefficient reached a maximum at slope of about 0.2 and then dropped when the slope became steeper for both SPPs. The RMSE of both SPPs presented slight decreasing trends at gently sloping area but became larger at steeper area. In general, TRMM 3B42 performed best within the ln(E) range of 4.0–5.0 and slope range of 0.1–0.3, but for PERSIANN, the dependence on elevation and slope were trivial at lower elevation and gently sloping area. At higher elevation (ln(E)>5.0) and steeper area (slope>0.3), the validity of both SPPs decreased with increasing elevation and slope. Similar conclusions were also indicated by Barros et al. (2006), which found that the TRMM’s precipitation radar has difficulties in detecting precipitation at high elevations. 5 Discussions The previous section presented the error statistics, which showed that the different satellite rainfall products have very different strengths and weaknesses under different rainfall intensities and in different seasons. Possible sources of errors may be associated with the effects of different sensors, topographies, and retrieval algorithms used in the rainfall estimates (Beighley et al. 2011). Poyang Lake basin consists of mountainous and hilly areas, where the complex topography could cause strong scattering signals in the microwave region, especially at cold land surfaces and ice-covered or snow-covered areas (Huffman et al. 2007; Scheel et al. 2011) and also a strong influence on TB and its polarization property with varying snow cover conditions, depending on exposure and the altitude in mountainous terrain (Amlien 2008; Scheel et al. 2011). Mountainous regions have relatively warm clouds, and the satellite sensors may not detect the rainfall from the warm clouds as the cloud tops would be too warm for IR thresholds to discriminate between raining and nonraining clouds Fig. 9 Scatter plots of the correlation coefficient and RMSE versus slope over the Poyang Lake basin 726 (Hong et al. 2007; Dinku et al. 2008; Bitew and Gebremichael 2010). Moreover, clouds over mountainous area could produce heavy rainfall without much ice aloft in PMW algorithms (Dinku et al. 2010). However, the sensors could accurately detect rainfall from the deep convection, as Fig. 5b shows a better performance in the summer season with the smaller BIAS (almost 0 in July) for PERSIANN data. Zhou et al. (2008) also gained similar conclusions that rainfall is more convective with higher rainfall intensities during the warm season and could be accurately estimated in satellite precipitation products. But, on the other hand, the heavy rainfall may cause signal attenuation which is significant and most frequently encountered (Villarini and Krajewski 2010). This is a possible explanation for the bad performances under higher rainfall intensity for both SPPs (Fig. 3). Additionally, although the topography obviously influences the accuracy of satellite products, the retrieval algorithm may significantly dominate the contributions of satellite error sources for high-resolution estimates (Yan and Gebremichael 2009; AghaKouchak et al. 2009). The current global algorithms estimate precipitation indirectly from the TB at the cloud top (Levizzani and Amorati 2002) and do not consider the altitude of the object and the sub-cloud evaporation (Scheel et al. 2011; Dinku et al. 2011), which significantly affect the retrieval accuracy of precipitation (Petty 2001). At the same time, further challenges arise from the processing scheme for microwave and IR data (Scheel et al. 2011). The definition of the underlying surface should satisfy the interpretation of the measured microwave signal and globally applied algorithms need to cope with highly heterogeneous terrain with varying TBs (Scheel et al. 2011). Furthermore, it is indispensable to calibrate the retrieval algorithms using locally available rain gauge observations, which is not just for the selection of appropriate temperature thresholds but also involves determining the other relevant calibration parameters (Dinku et al. 2008, 2011). Local calibration could be one of the most potent approaches to alleviate the satellite precipitation errors. Certainly, the current research is only limited to evaluate and compare the SPPs at the daily scale. The sub-daily (or three hourly) precipitation data are not involved due to the temporal scale limitation of rain gauge data, although they are more critical to drive the flood warning models and decision-making systems in Poyang Lake basin. Moreover, the effects of topography are more complex, as it includes other factors than just elevation and slope, i.e., the orientation of the slope with respect to wind direction at a given time, and geographical location of the slopes (Dinku et al. 2008). The current research is limited to describing the roles X. Li et al. that elevation and slope might have played in the performance of satellite rainfall products; however, the physical mechanisms are not addressed adequately here. So, extensive efforts on the evaluation of satellite-based products and thorough understanding of the errors in satellite rainfall need to continue in Poyang Lake basin as well as in other regions. 6 Conclusions This paper evaluated and compared two widely used highresolution satellite-based precipitation data (TRMM 3B42 V6 and PERSIANN) with rain gauge data in Poyang Lake basin and investigated their spatial and temporal characteristics, including their relationship with evaluation and slope. It is concluded that: & & & & & TRMM 3B42 and PERSIANN were better suited to determining rain occurrence frequency than to determining the rainfall amount. In the study region, the former slightly underestimated the rainfall amount contribution rates in middle rainfall class ranges (3 mm/day<rainfall≤25 mm/day) but overestimated it in the heavy rainfall class (> 50 mm/day), and the opposite trend was observed for PERSIANN. The temporal distribution characteristics of monthly rainfall were correctly described by both SPPs, and greater RMSEs were mainly observed between April and June. PERSIANN performed much worse in the winter season with the BIAS value being as much as −80 %, while TRMM 3B42 gained better accuracy in the winter season than in the summer season. TRMM 3B42 had better performance in estimating the frequency and locality of precipitation occurrence and had potential for useful application in regions where rain gauge observations were sparse or of bad quality. Shortcomings of TRMM 3B42 are as follows: it usually underestimates the rainy days and overestimates the average rainfall and intense rain events, which may reduce the accuracy of land surface hydrological processes simulation or flood forecasting. PERSIANN tended to suffer from a large systematic underestimation of rainfall, and during high precipitation events, the occurrence frequency and rainfall amount were underestimated greatly, which clearly revealed that IR-based rainfall algorithms had major limitations in reproducing rainfall fields in the Poyang Lake basin. The performances of both SPPs had clear dependence on elevation and slope and their relationships can be fitted using quadratic equations. TRMM 3B42 and PERSIANN performed best at a gently rolling landscape and the accuracy would decrease at higher elevation or steeper area. Assessing the performance of satellite-based precipitation products These conclusions indicate that efforts are necessary to further improve the current algorithms to reduce false alarms and missed precipitation and capture the heavy rain events correctly. Especially for PERSIANN, it is indispensable to incorporate additional information, such as relative humidity (Janowiak et al. 2001, 2004) and/or rain gauge data (Thorne et al. 2001), into the IR-based algorithms to improve the accuracy of rainfall estimates. On the other hand, it is an exigent need to develop and improve the adjustment procedure of hydrological models and flood warning models to advance the utilization of satellite-based precipitation data in practical applications (i.e., flood forecasting and warning). Acknowledgments This work is jointly funded by the National Basic Research Program of China (973 Program) (2012CB417003 and 2012CB956103-5), the National Natural Science Foundation of China (41101024), and the Science Foundation of Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences (NIGLAS2012135001 and NIGLAS2010XK02). The authors are grateful to the anonymous reviewers and the editor who helped in improving the quality of the original manuscript and Dr. Qing Zhu from Nanjing Institute of Geography and Limnology, CAS for providing valuable improvements to the earlier manuscript. Thanks also to Dr. Jian Liu and Dr. Yuanbo Liu from Nanjing Institute of Geography and Limnology, CAS for providing daily rain gauge data in Poyang Lake basin. References AghaKouchak A, Behrangi A, Sorooshian S, Hsu K, Amitai E (2011) Evaluation of satellite-retrieved extreme precipitation rates across the central United States. J Geophys Res 116:D02115. doi:10.1029/2010JD014741 AghaKouchak A, Nasrollahi N, Habib E (2009) Accounting for uncertainties of the TRMM satellite estimates. Remote Sens 1:606– 619 Amlien J (2008) Remote sensing of snow with passive microwave radiometers: a review of current algorithms. Report no. 1019. Norsk Regnesentral, Oslo, pp 1–52 Aonashi K, Awaka J, Hirose M, Kozu T, Kubota T, Liu G, Shige S, Kida S, Seto S, Takahashi N, Takayabu YN (2009) GSMaP passive microwave precipitation retrieval algorithm: algorithm description and validation. J Meteorol Soc Jap 87A:119–136 Artan G, Gadain H, Smith JL, Asante K, Bandaragoda CJ, Verdin JP (2007) Adequacy of satellite derived rainfall data for stream flow modeling. Nat Hazards 43(2):167–185 Asadullah A, Mcintyre N, Kigobe M (2008) Evaluation of five satellite products for estimation of rainfall over Uganda. Hydrolog Sci J 53 (6):1137–1150 Barros AP, Chiao S, Lang TJ, Burbank D, Putkonen J (2006) From weather to climate—seasonal and interannual variability of storms and implications for erosion process in the Himalaya. Geological Society of America Special Paper 398, Penrose Conference Series, Boulder, pp 17–38 Behrangi A, Khakbaz B, Jaw TC, AghaKouchak A, Hsu K, Sorooshian S (2011) Hydrologic evaluation of satellite precipitation products over a mid-size basin. J Hydrol 397:225–237 727 Beighley RE, Ray RL, He Y, Lee H, Schaller L, Andreadis KM, Durand M, Alsdorf DE, Shum CK (2011) Comparing satellite derived precipitation datasets using the Hillslope River Routing (HRR) model in the Congo River Basin. Hydrol Process 25:3216–3229 Bitew MM, Gebremichael M (2010) Evaluation through independent measurements: complex terrain and humid tropical region in Ethiopia. In: Gebremichael M, Hossain F (eds) Satellite rainfall applications for surface hydrology. Springer Science+Business Media B.V., Dordrecht, pp 205–214 Deyzel ITH, Pegram GGS, Visser PJM, Dicks D (2004) Spatial interpolation and mapping of rainfall (SIMAR) 2: radar and satellite products. WRC report no. 1152/1/04. Water Research Commission, Pretoria Dinku T, Anagnostou EN (2006) TRMM calibration of SSM/I algorithm for overland rainfall estimation. J Appl Meteorol Climatol 45(6):875–886 Dinku T, Ceccato P, Connor SJ (2011) Challenges of satellite rainfall estimation over mountainous and arid parts of east Africa. Int J Remote Sens 32(21):5965–5979 Dinku T, Chidzambwa S, Ceccato P, Connor SJ, Ropelewski CF (2008) Validation of high-resolution satellite rainfall products over complex terrain. Int J Remote Sens 29(14):4097–4110 Dinku T, Connor SJ, Ceccato P (2010) Comparison of CMORPH and TRMM-3B42 over mountainous regions of Africa and South America. In: Gebremichael M, Hossain F (eds) Satellite rainfall applications for surface hydrology. Springer Science + Business Media B.V., Dordrecht, pp 193–204 Ebert EE, Janowiak JE, Kidd C (2007) Comparison of near-real-time precipitation estimates from satellite observations and numerical models. B Am Meteorol Soc 88:47–64 Ferraro RR, Marks GF (1995) The development of SSM/I rain-rate retrieval algorithms using ground-based radar measurements. J Atm Oce Tech 12:755–770 Ghile Y, Schulze R, Brown C (2010) Evaluating the performance of ground-based and remotely sensed near real-time rainfall fields from a hydrological perspective. Hydrolog Sci J 55(4):497–511 Gottschalck J, Meng J, Rodell M, Houser P (2005) Analysis of multiple precipitation products and preliminary assessment of their impact on global land data assimilation system land surface states. J Hydrometeorol 6(5):573–598 Guo H, Hu Q, Jiang T (2008) Annual and seasonal streamflow responses to climate and land-cover changes in the Poyang Lake basin, China. J Hydrol 355:106–122 Han WS, Steven JB, Shepherd JM (2011) Assessment of satellitebased rainfall estimates in urban areas in different geographic and climatic regions. Nat Hazards 56:733–747 He X, Vejen F, Stisen S, Sonnenborg TO, Jensen KH (2011) An operational weather radar-based quantitative precipitation estimation and its application in catchment water resources modeling. Vadose Zone J 10(1):8–24 Hirpa FA, Gebremichael M, Hopson T (2010) Evaluation of highresolution satellite precipitation products over very complex terrain in Ethiopia. J Appl Meteorol Clim 49:1044–1051 Hong Y, Gochis D, Cheng J, Hsu K, Sorooshian S (2007) Evaluation of PERSIANN-CCS rainfall measurement using the NAME event rain gauge network. J Hydrometeorol 8(3):469–482 Hsu KL, Gao XG, Sorooshian S, Gupta HV (1997) Precipitation estimation from remotely sensed information using artificial neural networks. J Appl Meteorol 36:1176–1190 Hsu KL, Gupta HV, Gao X, Sorooshian S (1999) Estimation of physical variables from multi-channel remotely sensed imagery using a neural network: application to rainfall estimation. Water Resour Res 35(5):1605–1618 Hsu KL, Sorooshian S (2008) Satellite-based precipitation measurement using PERSIANN system. In: Sorooshian S et al (eds) 728 Hydrological modeling and the water cycle—coupling the atmospheric and hydrologic models. Springer, Berlin, pp 27–48 Hu Q, Feng S, Guo H, Jiang T (2007) Interactions of the Yangtze River flow and hydrologic processes of the Poyang Lake, China. J Hydrol 347:90–100 Huffman GJ, Adler RF, Rudolf B, Schneider U, Kehn PR (1995) Global precipitation estimates based on a technique for combining satellite-based estimates, rain gauge analysis, and NWP model precipitation estimation. J Climate 8:1284–1295 Huffman GJ, Adler RF, Arkin P, Chang A, Ferraro R, Gruber A, Janowiak J, McNab A, Rudolf B, Schneider B (1997) The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. B Am Meteorol Soc 78(1):5–20 Huffman GJ, Adler RF, Bolvin DT, Gu G, Nelkin EJ, Bowman KP, Hong Y, Stocker EF, Wolff DB (2007) The TRMM multi-satellite precipitation analysis (TMPA): quasiglobal, multiyear, combined sensor precipitation estimates at fine scales. J Hydrometeorol 8 (1):38–55 Hughes DA (2006) Comparison of satellite rainfall data with observations from gauging station networks. J Hydrol 327:399–410 Janowiak J, Joyce R, Yarosh Y (2001) A real-time global halfhourly pixel-resolution infrared dataset and its applications. B Am Meteorol Soc 82:205–217 Janowiak J E, Xie P, Joyce RJ, Chen M, Yarosh Y (2004) Validation of satellite-derived rainfall estimates and numerical model forecasts of precipitation over the United States. Proceeding of 29th Annual Climate Diagnostics and Prediction Workshop, NOAA, Madison, pp 21–26 Jiang H, Halverson JB, Zipser EJ (2008) Influence of environmental moisture on TRMM-derived tropical cyclone precipitation over land and ocean. Geophys Res Lett 35(17), L17806 Joyce RJ, Janowiak JE, Arkin PA, Xie P (2004) CMORPH: a method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J Hydrometeorol 5:487–503 Koo MS, Hong SY, Kim J (2009) An evaluation of the tropical rainfall measuring mission (TRMM) multi-satellite precipitation analysis (TMPA) data over South Korea. Asia-Pacific J Atmos Sci 45 (3):265–282 Krajewski WF, Smith JA (2002) Radar hydrology: rainfall estimation. Adv Water Resour 25:1387–1394 Kubota T, Shige S, Hashizume H, Aonashi K, Takahashi N, Seto S, Hirose M, Takayabu YN, Nakagawa K, Iwanami K, Ushio T, Kachi M, Okamoto K (2007) Global precipitation map using satelliteborne microwave radiometers by the GSMaP project: production and validation. IEEE Trans Geosci Remote Sens 45:2259–2275 Kubota T, Ushio T, Shige S, Kida S, Kachi M, Okamoto K (2009) Verification of high-resolution satellite-based rainfall estimates around Japan using a gauge-calibrated ground-radar dataset. J Meteorol Soc Jap 87A:203–222 Kummerow C, Barnes W, Koju T, Shiue J, Simpson J (1998) The tropical rainfall measuring mission (TRMM) sensor package. J Atmos Ocean Tech 15:809–817 Kummerow C, Hong Y, Olson WS, Yang S, Adler RF, McCollum J, Ferraro R, Petty G, Shin DB, Wilheit TT (2001) The evolution of the Goddard profiling algorithm (GPROF) for rainfall estimation from passive microwave sensors. J Appl Meteor 40:1801–1820 Layberry R, Kniveton DR, Todd MC, Kidd C, Bellerby TJ (2006) Daily precipitation over Southern Africa: a new resource for climate studies. J Hydrometeorol 7:149–159 Levizzani V, Amorati R (2002) A review of satellite-based rainfall estimation methods: a look back and a perspective. Proceedings of the 2000 EUMETSAT Meteorological Satellite Data User’s Conference, 29 May–2 June 2000, Bologna, Italy, pp 344–353 X. Li et al. Li L, Ngongondo CS, Xu CY, Gong L (2013) Comparison of the global TRMM and WFD precipitation datasets in driving a large-scale hydrological model in Southern Africa. Hydrol Res. doi:10.2166/nh.2012.175 Li XH, Zhang Q, Xu CY (2012) Suitability of the TRMM satellite rainfalls in driving a distributed hydrological model for water balance computations in Xinjiang catchment, Poyang Lake basin. J Hydrol 426–427:28–38 Nair S, Srinivasan G, Nemani R (2009) Evaluation of multi-satellite TRMM derived rainfall estimates over a western state of India. J Meteorol Soc Jap 87(6):927–939 Nesbitt SW, Gochis DJ, Lang TJ (2008) The diurnal cycle of clouds and precipitation along the sierra madre occidental observed during NAME-2004: implications for warm season precipitation estimation in complex terrain. J Hydrometeorol 9(4):728–743 Pegram GGS, Deyzel ITH, Sinclair S, Visser P, Terblanche D, Green GC (2004) Daily mapping of 24 hr rainfall at pixel scale over South Africa using satellite, radar and raingauge data. Proceeding of 2nd International Precipitation Working Group (IPWG) Workshop, Naval Research Laboratory, Monterey Petty GW (2001) Physical and microwave radiative properties of precipitation clouds. Part II: a parametric 1D rain-cloud model for use in microwave radiative transfer simulations. J Appl Meteorol 40:2115–2129 Piccolo F, Chirico GB (2005) Sampling errors in rainfall measurements by weather radar. Adv Geosci 2:151–155 Romilly TG, Gebremichael M (2011) Evaluation of satellite rainfall estimates over Ethiopian river basins. Hydrol Earth Syst Sci 15:1505–1514 Samaniego L, Kumar R, Jackisch C (2012) Predictions in a data-sparse region using a regionalized grid-based hydrologic model driven by remotely sensed data. Hydrol Res 42(5):338–355 Sapiano MRP, Arkin PA (2009) An intercomparison and validation of high-resolution satellite precipitation estimates with 3-hourly gauge data. J Hydrometeorol 10(1):149–166 Sawunyama T, Hughes DA (2008) Application of satellite-derived rainfall estimates to extend water resource simulation modelling in South Africa. Water SA 34:1–9 Scheel MLM, Rohrer M, Huggel CH, Villar DS, Silvestre E, Huffman GJ (2011) Evaluation of TRMM multi-satellite precipitation analysis (TMPA) performance in the central Andes region and its dependency on spatial and temporal resolution. Hydrol Earth Syst Sci 15:2649–2663 Schulze RE (2006) South african atlas of climatology and agrohydrology. WRC report no. 1489/1/06. Water Research Commission, Pretoria Shrestha MS, Takara K, Kubota T, Bajracharya SR (2011) Verification of GSMaP rainfall estimates over the central Himalaya. Ann J Hydraulic Eng, JSCE 55:37–42 Smith JA, Baeck ML, Meierdiercks KL, Miller AJ, Krajewski WF (2007) Radar rainfall estimation for flash flood forecasting in small urban watersheds. Adv Water Resour 30:2087–2097 Sorooshian S, Hsu KL, Gao X, Gupta HV, Imam B, Braithwaite D (2000) Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. B Am Meteorol Soc 81:2035–2046 Sorooshian S, Lawford R, Try P, Rossow W, Roads J, Polcher J, Sommeria G, Schiffer R (2005) Water and energy cycles: investigating the links. WMO Bull 54(2):58–64 Stisen S, Sandholt I (2010) Evaluation of remote-sensing-based rainfall products through predictive capability in hydrological runoff modeling. Hydrol Process 24(7):879–891 Su FG, Hong Y, Lettenmaier DP (2008) Evaluation of TRMM multisatellite precipitation analysis (TMPA) and its utility in hydrologic prediction in La Plata Basin. J Hydrometeorol 9(4):622–640 Sun X, Mein RG, Keenan TD, Elliott JF (2000) Flood estimation using radar and raingauge data. J Hydrol 239:4–18 Assessing the performance of satellite-based precipitation products Thorne V, Coakley P, Grimes D, Dugdale G (2001) Comparison of TAMSAT and CPC rainfall estimates with rainfall, for southern Africa. Int J Remote Sens 22(10):1951–1974 Tian YD, Peters-Lidard CD (2007) Systematic anomalies over inland water bodies in satellite-based precipitation estimates. Geophys Res Lett 34(14), L14403 Tian YD, Peters-Lidard CD, Choudhury BJ, Garcia M (2007) Multitemporal analysis of TRMM-based satellite precipitation products for land data assimilation applications. J Hydrometeorol 8(6):1165–1183 Turk FJ, Rohaly GD, Hawkins J, Smith EA, Marzano FS, Mugnai A, Levizzani V (2000) Meteorological applications of precipitation estimation from combined SSM/I, TRMM and infrared geostationary satellite data. In: Pampaloni P, Paloscia S (eds) Microwave radiometry and remote sensing of the Earth’s surface and atmosphere. VSP International Science Publishers, Zeist, pp 353–363 Ushio T, Sasashige K, Kubota T, Shige S, Okamoto K, Aonashi K, Inoue T, Takahashi N, Iguchi T, Kachi M, Oki R, Morimoto T, Kawasaki Z (2009) A kalman filter approach to the global satellitemapping of precipitation (GSMaP) from combined passive microwave and infrared radiometric data. J Meteorol Soc Jap 87A:137–151 Vila DA, Luis GG, Toll DL, Rozante JR (2009) Statistical evaluation of combined daily gauge observations and rainfall satellite estimates over continental South America. J Hydrometeor 10:533–543 Villarini G, Krajewski WF (2010) Review of the different sources of uncertainty in single polarization radar-based estimates of rainfall. Surv Geophys 31:107–129 729 Ward E, Buytaert W, Peaver L, Wheater H (2011) Evaluation of precipitation products over complex mountainous terrain: a water resources perspective. Adv Water Resour 34:1222–1231 Wilks DS (1995) Statistical methods in the atmospheric sciences. Academic, San Diego Wilks DS (2006) Statistical methods in the atmospheric sciences, 2nd edn. Academic, Burlington Xu DL, Xiong M, Zhang J (2001) Analysis of hydrological characteristic of Poyang Lake. Yangtze River 32(2):21–23 (in Chinese) Yamamoto MK, Ueno K, Nakamura K (2011) Comparison of satellite precipitation products with rain gauge data for the Khumb region, Nepal Himalayas. J Meteorol Soc Jap 89(6):597–610 Yan J, Gebremichael M (2009) Estimating actual rainfall from satellite rainfall products. Atmosph Res 92:481–488 Yong B, Ren LL, Hong Y, Wang JH, Gourley JJ, Jiang SH, Chen X, Wang W (2010) Hydrologic evaluation of multisatellite precipitation analysis standard precipitation products in basins beyond its inclined latitude band: A case study in Laohahe basin, China. Water Resour Res 46: W07542. doi:10.1029/2009WR008965 Yu Z, Yu H, Chen P (2009) Verification of tropical cyclone-related satellite precipitation estimates in mainland China. J Appl Meteorol Clim 48(11):2227–2241 Zhou T, Yu R, Chen H, Dai A, Pan Y (2008) Summer precipitation frequency, intensity and diurnal cycle over China: a comparison of satellite data with rain gauge observations. J Clim 21(16):3997–4040